You cannot draw a straight line from a ChatGPT mention to a checkout. So you build a proxy chain instead: share of voice, then a sales proxy like Best Sellers Rank, then revenue over a longer window. Here is how to make the case honestly, and why BSR beats revenue as the near-term baseline.
From Share of Voice to Sales: Building the AI-Visibility-to-Revenue Proof Chain
By Stephen Honight, Founder of Lmo7
Every brand I speak to right now arrives at the same question. They can see AI is changing how people find and choose products. What they cannot see is whether the work to show up in AI is moving sales. Until they can, the budget conversation stays stuck.
This is the most common thing I hear in client calls at the moment. A brand runs a Share-of-Model baseline, sees it is under-represented against competitors, agrees the work matters, and then a finance or leadership question lands. If we invest in this, how will we know it worked? It is a fair question. It is also the hardest one to answer cleanly, because the honest answer is that you cannot draw a straight line from a ChatGPT mention to a checkout.
My sense is that most of the confusion here comes from expecting AI visibility to behave like paid search. It does not. So before you can prove anything, you have to be clear about what is actually measurable and what has to be modelled instead.
Why the straight line does not exist
Start with the mechanism, because the measurement problem falls out of it.
When someone asks ChatGPT or Gemini or Rufus what to buy, the model gives an answer that names some brands and leaves others out. If your brand gets named, you have influenced the consideration set. But the person does not usually click a tracked link and buy in the same session. They read the answer, form a view, and go somewhere else to purchase. Amazon. Your D2C site. A retailer. Sometimes days later.
The referral data captures a sliver of this. When AI platforms do pass a click through, you can see it in analytics, and that number is real and worth watching. But it undercounts the actual influence badly, because most of the effect is people acting on a recommendation without ever clicking the source. The model shaped the decision and then disappeared from the trail.
So the visibility work is real and the sales effect is real, but the connective tissue between them is missing from the data. That is the gap every measurement framework has to deal with. You are not going to close it with a cleverer attribution model. You close it by building a chain of proxies that each move in a believable sequence.
The proxy chain
The way I would frame it is three links, moving from what you can measure directly to what you actually care about.
The first link is share of voice. This is the one thing you can measure cleanly and repeatedly in AI. Across a defined set of buyer prompts, how often does the model name you, in what position, with what sentiment, compared to your competitors. Run the same prompt set at two points in time and you have a clear before and after. This is the input you control and the one that moves first when the content and authority work lands.
The second link is a sales proxy. Something closer to revenue that still moves fast enough to read within an engagement. On Amazon that proxy is Best Sellers Rank. BSR reflects sales velocity relative to the category, updates constantly, and does not require anyone to share their P&L. It sits between visibility and revenue and it moves in weeks, not quarters.
The third link is revenue itself, read over a longer window. This is the number the business ultimately cares about, but it is also the noisiest and the slowest, and it is influenced by a dozen things that have nothing to do with AI. You look at it, but you look at it last and with the most caution.
The argument you are making to a client is not “AI visibility caused this revenue”. It is “AI visibility went up, then the sales proxy went up in the same window, then revenue followed over a longer period, and the sequence is consistent with the mechanism”. That is a proxy chain, not an attribution claim. It is honest, and in my experience it is more persuasive than a false precision number that anyone commercially literate will see straight through.
Why BSR beats revenue as the near-term baseline
Here is the part that keeps coming up in client work, and it is worth spelling out because it changes how you set the baseline.
If you anchor your proof on revenue, you inherit all the noise in revenue. The biggest source of that noise for a consumer brand on Amazon is promotional distortion. Prime Day, Black Friday, a Subscribe and Save push, a lightning deal, a competitor going out of stock. Any of these can swing revenue in a month by more than your AI visibility work will, and none of them tell you anything about whether the visibility work is landing.
Best Sellers Rank is not immune to this, but it is cleaner, because it is a relative measure. BSR reflects your sales velocity against the rest of the category. When the whole category lifts during Prime Day, your rank does not necessarily move, because everyone rose together. That relative framing strips out a lot of the promotional noise that makes raw revenue such a difficult baseline in the near term.
So my recommendation would be to set BSR as the near-term baseline metric and treat revenue as the longer-horizon confirmation. Take a clean baseline month before the work starts. Avoid anchoring on a month that contains a major promotional event, because a Prime Day baseline flatters or distorts everything that follows. Then compare against an equally clean month later in the engagement. Baseline month versus a clean later month, on BSR, is a far more readable signal than revenue month over month.
This is not a reason to ignore revenue. It is a reason to sequence it correctly. BSR gives you an early, defensible read. Revenue gives you the eventual commercial confirmation once enough time has passed for the effect to accumulate.
Set a concrete target, not “get more visible”
One thing I have learned watching these programmes run is that a vague goal produces a vague result. “Get more visible in AI” is not a target. It gives you nothing to measure against and nothing to hold the work to.
A concrete share of voice target does the job. Something like moving from 8% share of voice in a defined category prompt set to 12% over a quarter. That is specific enough to plan against, to report against, and to know whether you hit. It also forces the useful conversation up front about which prompts count, which competitors you are measured against, and what a realistic move looks like given your starting authority.
We do this with the spirits and consumer health brands we work with, and the discipline of naming a number changes the engagement. It turns “are we doing better” into “did we move from here to there”, which is a question you can actually answer.
The other benefit is that a share of voice target is honest about the layer you control. You control content, structure, claims coverage and eventually authority. You do not control the checkout. Setting the target on the thing you influence, then tracking the proxy and the revenue behind it, keeps the promise aligned with what the work can actually do.
The honest limit
I would not pitch any of this without being upfront about two things, because overclaiming here damages trust fast.
The first is that this is correlation, not attribution. The proxy chain shows a believable sequence. It does not prove causation in the way a controlled test would. Where you can run a genuine control versus test, do it, because that is the strongest evidence available. Most of the time in AI visibility you cannot cleanly isolate the variable, so you are reading a consistent pattern across the chain and being honest that it is a pattern.
The second is that it takes months. Revenue attribution in AI is still limited today, and it will improve as ChatGPT and Gemini build out their ad and measurement infrastructure over the next six to nine months. Until then, the effect accumulates slowly. Domain authority, which is the biggest lever on whether models cite you at all, is a slow build by nature. Anyone promising a clean revenue line from AI visibility inside a few weeks is selling something I would not buy.
The credible position is that the visibility layer is measurable now, the sales proxy is readable in weeks, and the revenue confirmation lands over quarters. Say that plainly and the client trusts the rest of the numbers more, not less.
What good looks like
The programmes that have produced real movement did it by holding this chain steady over time rather than chasing a single headline number.
With Trip Drinks we ran a 60-day AI search engagement focused on site signals, content upgrades and citation-source work. Over that window their average position across the major models moved from 7th to 3rd, and AI referral traffic rose 33%. The visibility link moved first, exactly as the mechanism predicts, and the referral signal followed. That is the front of the proof chain working in practice.
With Haleon we ran Share-of-Model analysis across a set of their brands. Voltarol came back with a 100% mention rate and the top average position in its category. That is a visibility baseline you can build a target and a proof chain on, because you know precisely where the brand stands before any further work starts. The baseline is the thing that makes everything downstream measurable.
Neither of these is a claim that AI visibility printed a specific revenue number. They are examples of the measurable links moving in the right order, which is the honest version of proof and the one that holds up.
So what should you do next
If measurement is the thing blocking your AI search budget, here is where I would start, in priority order.
Set your baseline before you touch anything. Pick a clean, non-promotional month. Capture share of voice across a defined buyer prompt set and capture BSR on your priority ASINs. You cannot prove movement without a credible starting point, and a baseline taken during Prime Day will mislead you for the rest of the year. If you want the data layer to run this yourself, our DaaS tier gives your team direct access to the source data across Amazon and the AI surfaces, plus the upskilling to read it. That is the right entry point if you have the capacity to action the work internally.
Name a concrete share of voice target and a comparison window. Move from X% to Y% over a quarter, measured baseline month versus a clean later month. This turns the programme into something you can hold to a number. For a single brand that wants us to run the track, diagnose, fix and re-track loop, the Challenger tier is built for exactly this cadence.
Report the proxy chain, not a single vanity metric. Show share of voice, then BSR, then revenue over a longer horizon, and be explicit about what is measured and what is modelled. For a multi-brand portfolio that needs this proof chain built and tracked across brands for internal stakeholders, that is Enterprise work, where the value is the comparative view and the alignment it creates.
The brands that win the budget argument are not the ones with the boldest attribution claim. They are the ones who set an honest baseline, name a target, and show the links moving in the right order over time. If you want help building that proof chain for your brand, that is the conversation to have.
Lmo7 helps consumer brands win in AI-powered discovery and agentic commerce across Amazon and D2C. We work with brands including Trip Drinks, Veloforte, Brown-Forman, Haleon, Pelotan and Symprove on AI search visibility, Amazon and Rufus optimisation, and the measurement layer that connects the two. If you want to know where your brand stands in AI-powered discovery, and how to prove the work is moving sales, get in touch.